MODEL SOVEREIGNTY RADAR Β· V12 Β· EOSE LABS Β· DAY 89
MODEL RADAR
WHAT LEAVES msi01 Β· γ₁ = 14.134725141734693
LOCAL β€” never leaves
ALLIED β€” Canada/EU
EXTERNAL β€” US servers
SENSITIVE β€” CHINA ⚠️
SECTION 2 β€” SOVEREIGNTY TABLE
PROVIDER CLASS AGENTS ASSIGNED DATA LEAVES COST/CALL STATUS
mal/qwen2.5:32b 🟒 LOCAL defaults, fallbacks Never $0 ● ACTIVE
ollama-local 🟒 LOCAL embed, memory Never $0 ● ACTIVE
lounge/qwen2.5:32b 🟒 LOCAL local :11434 Never $0 ● ACTIVE
anthropic/claude-sonnet-4-6 🟠 EXTERNAL main session US servers $3/$15 per 1M ● ACTIVE
openai/gpt-4.1 🟠 EXTERNAL officer, bob, bosun, signals, rick, helmsman, lucien, luffy + 15 more US servers varies ● ACTIVE
cohere/command-a 🟑 ALLIED (inactive) Canada low β—‹ STANDBY
mistral/mistral-large 🟑 ALLIED (inactive) France/EU low β—‹ STANDBY
gemini/gemini-2.0-flash 🟠 EXTERNAL ext fallback US/Google low β—‘ FALLBACK
deepseek/deepseek-chat πŸ”΄ SENSITIVE (inactive) ⚠️ CHINA low ⚠ RESTRICTED
xai/grok-3 🟠 EXTERNAL (inactive) US/xAI low β—‹ STANDBY
perplexity/sonar 🟠 EXTERNAL (inactive) US low β—‹ STANDBY
SECTION 3 — DEBT→CLOAK FEED
WPA < 50%
Silo healthy
β†’
NOMINAL β€” external routing allowed
WPA 50–84%
Silo under pressure
β†’
WATCH β€” flag sensitive data, prefer local
WPA β‰₯ 84.8%
γ₁ Γ— 6 = BREAK
β†’
BREAK β€” FORCE LOCAL β€” no external routing
── BFE LAYER β†’ CLOAK RULE ──
L1–L2 OPS
β†’
external OK β€” low exposure
L3 OPS
β†’
prefer local β€” concurrency data sensitive
L4 OPS
β†’
local strongly preferred β€” composition patterns = IP
L5 OPS
β†’
LOCAL ONLY β€” no exceptions β€” liability data never leaves
SECTION 4 β€” WHAT'S AT RISK
πŸ”΄
DeepSeek Restriction
Configured but restricted. Chinese servers. Never use for: DCJ numbers, moat strategy, patient patterns, fleet IP, architecture details, Amani's work. Safe for public knowledge queries only.
RESTRICTED · sovereignty_class = SENSITIVE · CHINA ⚠️
🟠
Anthropic Scan Risk
Anthropic confirmed scanning git commits for agent harness terms. Main session flows to Anthropic. Everything discussed is logged. For highly sensitive IP: use local Qwen.
P1 Β· US servers Β· EXTERNAL class Β· git commit scanning confirmed
🟠
OpenAI Exposure β€” 15+ Agents
bob, officer, bosun, signals, rick, helmsman, lucien, luffy + 15 more all default to gpt-4.1. Every subagent task spawned today hit OpenAI. Cost accumulates + IP exposure grows. Migration path: swap to mal/qwen2.5:32b for mechanical tasks.
P1 Β· $0.01–0.05 per call Β· US servers Β· migrate via phases 1–4
🟑
Cohere Advantage
Canadian servers. PIPEDA-governed. If you need an external model with the best data sovereignty in Canada, Cohere command-a is the right call. Currently standby.
ALLIED class Β· Canada Β· PIPEDA Β· currently STANDBY β€” recommend activating
SECTION 5 β€” THE MIGRATION PATH
PHASE 1 Β· NOW
Mechanical Subagents β†’ mal/qwen2.5:32b
builds, deploys, file ops β€” zero IP exposure, $0 cost, immediate
↓
PHASE 2 Β· SOON
Named Agents β†’ mal/qwen2.5:32b
bob, bosun, signals, rick migrate off gpt-4.1 β€” cuts daily OpenAI burn by ~60%
↓
PHASE 3 Β· V12
Main Session Fallback β†’ Local First
Anthropic only for reasoning-heavy tasks β€” local Qwen as default for everything else
↓
PHASE 4 Β· V12
BFE L4/L5 Operations β†’ FORCE LOCAL in Cloak Rules
Debt helix BREAK signal wired to cloak β€” high-WPA silos automatically go dark to external providers
πŸ’° SAVINGS ESTIMATE Β· today's session ran 6+ heavy subagents all hitting gpt-4.1 at ~$0.01–0.05 per call = $0.06–0.30 just today. At fleet scale across 7+ silos: significant. Phase 1+2 migration = ~$0.15–0.50/day saved.
γ₁ = 14.134725141734693 Β· MODEL SOVEREIGNTY RADAR Β· V12 Β· EOSE LABS INC Β· DAY 89 Β· WHAT LEAVES msi01